[R] Forcing the extrapolation of loess through the origin

[Ravi Varadhan]

Another approach would be to use the "weights" argument. So, for example,
you can use a weight of 100 for the point (0,0), and weights of 1 for the
rest of the data.

However, I would be wary of this, as I am not sure why you want to do this.


Here is an example:

 set.seed(123)

x <- sort(runif(100))

y <- 1.2 * x^2 - 0.5 * x + rnorm(100, sd=0.1)

y.new <- c(0, y)

x.new <- c(0, x)

obj <- loess(y ~ x, span=0.3)

plot(x, y)

lines(x, predict(obj), col=2, lty=2)

obj.new <- loess(y.new ~ x.new, span=0.3, weights=c(100, rep(1, 100)))

lines(x.new, predict(obj.new), col=4, lty=3)


Note that the constraint does "significantly" alter the loess smooth close
to the origin, as it should.

Ravi.

----------------------------------------------------------------------------
-------

Ravi Varadhan, Ph.D.

Assistant Professor, The Center on Aging and Health

Division of Geriatric Medicine and Gerontology 

Johns Hopkins University

Ph: (410) 502-2619

Fax: (410) 614-9625

Email: rvaradhan at jhmi.edu

Webpage:  http://www.jhsph.edu/agingandhealth/People/Faculty/Varadhan.html

 

----------------------------------------------------------------------------
--------


-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of Stavros Macrakis
Sent: Tuesday, April 14, 2009 2:15 PM
To: jimm-panse at gmx.de
Cc: r-help at r-project.org
Subject: Re: [R] Forcing the extrapolation of loess through the origin

On Tue, Apr 14, 2009 at 1:08 PM,  <jimm-panse at gmx.de> wrote:
> I'm fitting a line to my dataset. Later I want to predict missing values
that exceed the [min,max] interval of my empirical data, therefore I choose
surface="direct" for extrapolation.
>
> l1<-loess(y1~x1,span=0.1,data.frame(x=x1,y=y1),control=loess.control(s
> urface="direct"))
>
> In my application it is highly important that the fitted line intercepts
at the point of origin. Is it possible to do this in R?

Well, you could always add lots of artificial data points x=0, y=0 ..., like
this:

l1<-loess(y1~x1,span=0.1,data.frame(x=c(rep(0,100),x1),y=c(rep(0,100),y1)),c
ontrol=loess.control(surface="direct"))

which will eventually drive f(0) to near 0, but surely that will create
fitting artifacts.

          -s

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.